Testing the MAGNETO (Mood As Global Net Expected Trade-Off) model

Two views on how emotions impact decisions have been opposed across centuries. The Cartesian view suggests that a rational decision-maker must ignore emotions, whereas the Darwinian view suggests that emotions help the decision-maker with valuating options. Our lab has conducted studies that confront these two perspectives, applied to mood fluctuations. Compared to emotions, moods are affective states that are not tight to a particular trigger and that last long enough to influence unrelated decisions. Here, we adopt a definition of mood as being unidimensional: it varies from bad to good during normal fluctuations, or between depression and mania during pathological episodes. We recently developed a computational model dubbed MAGNETO (Pessiglione et al., 2023) that accounts for a rudimentary form of mood, which we might share with our mammalian ancestors. In this model, mood is simply the average of all possible action values, which are learned from their outcomes (obtained rewards and incurred costs). In turn, mood exerts a bias on decisions about foraging actions: good mood make foraging actions more likely, such that agents are willing to incur more costs (make efforts, take risks, wait for delays) and obtain more rewards (food, partner, shelter, etc.) Simulations show that agents equipped with this mood-related decision bias strive better than cold agents in environments where the different sources of rewards and costs are correlated (but not if they are independent). Therefore, both the Cartesian and Dawinian views are correct: depending on the environment, the impact of mood on decisions can be irrational or rational. We have validated the impact of mood on decisions in a series of behavioral and physiological experiments (Heerema et al., 2023; 2025). The aim of this project is to further test the validity of the complete model, including the generation of mood fluctuations in addition to their impact on decisions. More precisely, the reason why the mood-related decision bias may be adaptive is that it provides generalization across actions, saving the need for systematic trial and error testing. Thus, the idea is to assess this critical feature of the model, by setting up a new behavioral task. The task must include at least three actions, for which the contingencies (associations with risk and reward levels) oscillate across trials. Oscillations of actions A and B would be correlated, with high and low amplitudes, whereas oscillations of actions C would be anti-correlated, with low amplitude. The prediction of the model is that actions B and C should be engaged more frequently when the cost/benefit trade-off of action A is peaking, which would be adaptive for B but maladaptive for C. For the M2 internship, the project consists in developing the task, testing healthy volunteers and updating the computational model based on behavioral results. It could be pursued in a PhD internship that would aim at 1) elucidating the brain mechanisms underlying the mood-related decision bias using neuroimaging techniques, 2) specifying the distortions of the bias that may lead to catastrophic generalization in psychiatric conditions (notably mood disorders), 3) extending the computational model to incorporate more sophisticated inferences that may affect mood fluctuations in the human brain.